21.

Maxima and minima: II To see how to maximise and minimise a function on the boundary, lets conside a concrete example. Let K = { (x, y ) | x2 + y 2 2 }. Then K is compact. Let f : K R, be the function f (x, y ) = xy . Then f is continuous and so f achieves its maximum and minimum. I. Lets rst consider the interior points. Then f (x, y ) = (y, x), so that (0, 0) is the only critical point. The Hessian of f is Hf (x, y ) = 0 1 . 1 0

d1 = 0 and d2 = 1 = 0 so that (0, 0) is a saddle point. It follows that the maxima and minima of f are on the boundary, that is, the set of points C = { (x, y ) | x2 + y 2 = 2 }. II. Let g : R2 R be the function g (x, y ) = x2 + y 2 . Then the circle C is a level curve of g . The original problem asks to maximise and minimise f (x, y ) = xy subject to g (x, y ) = x2 + y 2 = 2.

One way to proceed is to use the second equation to eliminate a variable. The method of Lagrange multipliers does exactly the opposite. Instead of eliminating a variable we add one more variable, traditionally called . So now lets maximise and minimise h(x, y, ) = f (x, y ) (g (x, y ) 2) = xy (x2 + y 2 2). We nd the critical points of h(x, y, ): y = 2x x = 2y 2 = x2 + y 2 . First note that if x = 0 then y = 0 and x2 + y 2 = 0 = 2, impossible. So x = 0. Similarly one can check that y = 0 and = 0. Divide the
1

rst equation by the second: y x = , x y so that y 2 = x2 . As x2 + y 2 = 2 it follows that x2 = y 2 = 1. So x = 1 and y = 1. This gives four potential points (1, 1), (1, 1), (1, 1), (1, 1). Then the maximum value of f is 1, and this occurs at the rst and the last point. The minimum value of f is 1, and this occurs at the second and the third point. One can also try to parametrise the boundary: r(t) = 2(cos t, sin t). So we maximise the composition h : [0, 2 ] R, where h(t) = 2 cos t sin t. As I = [0, 2 ] is compact, h has a maximum and minimum on I . When h (t) = 0, we get cos2 t sin2 t = 0. Note that the LHS is cos 2t, so we want cos 2t = 0. It follows that 2t = /2 + 2m , so that t = /4, 3/4, 5/4, and 7/4. These give the four points we had before. What is the closest point to the origin on the surface F = { (x, y, z ) R3 | x 0, y 0, z 0, xyz = p }? So we want to minimise the distance to the origin on F . The rst trick is to minimise the square of the distance. In other words, we are trying to minimise f (x, y, z ) = x2 + y 2 + z 2 on the surface F = { (x, y, z ) R3 | x 0, y 0, z 0, xyz = p }. In words, given three numbers x , y 0 and z 0 whose product is p > 0, what is the minimum value of x2 + y 2 + z 2 ? Now F is closed but it is not bounded, so it is not even clear that the minimum exists. Lets use the method of Lagrange multipliers. Let h : R4 R, be the function h(x, y, z, ) = x2 + y 2 + z 2 (xyz p).
2

We look for the critical points of h: 2x = yz 2y = xz 2z = xy p = xyz. Once again, it is not possible for any of the variables to be zero. Taking the product of the rst three equations, we get 8(xyz ) = 3 (x2 y 2 z 2 ). So, dividing by xyz and using the last equation, we get 8 = 3 p, that is = 2 p1/3 .

Taking the product of the rst two equations, and dividing by xy , we get 4 = 2 z 2 , so that z = p1/3 . So h(x, y, z, ) has a critical point at 2 (x, y, z, ) = (p1/3 , p1/3 , p1/3 , 1/3 ). p We check that the point (x, y, z ) = (p1/3 , p1/3 , p1/3 ), is a minimum of x2 + y 2 + z 2 subject to the constraint xyz = p. At this point the sum of the squares is 3p2/3 . 1/3 Suppose that x 3p . Then the sum of the squares is at least 3p2/3 . Similarly if y 3p1/3 or z 3p1/3 . On the other hand, the set K = { (x, y, z ) R3 | x [0, 3p1/3 ], y [0, 3p1/3 ], z [0, 3p1/3 ], xyz = p }, is closed and bounded, so that f achieves it minimum on this set, which we have already decided is at (x, y, z ) = (p1/3 , p1/3 , p1/3 ), since f is larger on the boundary. Putting all of this together, the point (x, y, z ) = (p1/3 , p1/3 , p1/3 ),
3

is a point where the sum of the squares is a minimum. Here is another such problem. Find the closest point to the origin which also belongs to the cone x2 + y 2 = z 2 , and to the plane x + y + z = 3. As before, we minimise f (x, y, z ) = x2 + y 2 + z 2 subject to g1 (x, y, z ) = x2 + y 2 z 2 = 0 and g2 (x, y, z ) = x + y + z = 3. Introduce a new function, with two new variables 1 and 2 , h : R5 R, given by h(x, y, z, 1 , 2 ) = f (x, y, z ) 1 g1 (x, y, z ) 2 g2 (x, y, z ) = x2 + y 2 + z 2 1 (x2 + y 2 z 2 ) 2 (x + y + z 3). We nd the critical points of h: 2x = 21 x + 2 2y = 21 y + 2 2z = 21 z + 2 z 2 = x2 + y 2 3 = x + y + z. Suppose we substract the rst equation from the second: y x = 1 (y x). So either x = y or 1 = 1. Suppose x = y . Then 1 = 1 and 2 = 0. In this case z = z , so that z = 0. But then x2 + y 2 = 0 and so x = y = 0, which is not possible. It follows that x = y , in which case z = 2x and (2 2)x = 3. So 3 3(2 2) = x= . 2 2 2 This gives us two critical points: 3(2 2) 3(2 2) 3 2(2 2) P1 = ( , , ) 2 2 2 3(2 + 2) 3(2 + 2) 3 2(2 2) , , ). P2 = ( 2 2 2 Of the two, clearly the rst is closest to the origin.
4

To nish, we had better show that this point is the closest to the origin on the whole locus F = { (x, y, z ) R3 | x2 + y 2 = z 2 , x + y + z = 3 }. Let K = { (x, y, z ) F | x2 + y 2 + z 2 25 }. Then K is closed and bounded, whence compact. So f achieves its minimum somewhere on K , and so it must achieve its minimum at P = P1 . Clearly outside f is at least 25 on F \ K , and so f is a minimum at P1 on the whole of F .